|
Bugzilla – Full Text Bug Listing |
| Summary: | 2.5(.9) qsub does not seem to accept comma seperated -W argument | ||
|---|---|---|---|
| Product: | TORQUE | Reporter: | ramon.bastiaans |
| Component: | clients | Assignee: | Ken Nielson <knielson> |
| Status: | RESOLVED FIXED | ||
| Severity: | major | CC: | chris, rea+maui, torquedev |
| Priority: | P5 | ||
| Version: | 2.5.x | ||
| Hardware: | PC | ||
| OS: | Linux | ||
| Attachments: |
Fixes handling of multiple comma-separated arguments for -W
Fixes old regression: qsub -W won't accept directive like stagein=a@h:b,c@h:d |
||
(In reply to comment #0) > I believe 2.4 suffered from a similar bug, which was fixed in 2.4.9 (from > changelog): > > b - restore ability to parse -W x=geometry:{...,...} That was SVN commit 3774, "Earlier patch broke ability to parse geometry:{...} now fixed", or git commit 98d76c1a7f4da64fb38e2846a473baf9116023c1 for those following along at home with git-svn. > Is it possible that this bug was not fixed in 2.5 ? Or maybe this is an > completely different bug. Looks like the fix was never applied to 2.5, the 2.4 change still applies cleanly (at least to qsub.c, the CHANGELOG part fails of course).. samuel@eris:~/Downloads/Torque$ git show 98d76c1a7f4da64fb38e2846a473baf9116023c1 > /tmp/qsub.patch samuel@eris:~/Downloads/Torque$ cd branches/2.5-fixes/ samuel@eris:~/Downloads/Torque/branches/2.5-fixes$ patch --dry-run -p3 < /tmp/qsub.patch patching file CHANGELOG Hunk #1 FAILED at 28. 1 out of 1 hunk FAILED -- saving rejects to file CHANGELOG.rej patching file src/cmds/qsub.c Hunk #1 succeeded at 491 (offset 19 lines).
Created an attachment (id=101) [details]
Fixes handling of multiple comma-separated arguments for -W
I have the same issue for 2.5.10. The attached patch fixes the problem for me.
Please, try it (though I am not sure that this patch will work for the whole
2.5 line).
Created an attachment (id=102) [details] Fixes old regression: qsub -W won't accept directive like stagein=a@h:b,c@h:d And another patch that fixes the long-standing regression in the qsub behaviour: as the manual says, -W stagein has the form 'stagein=file_list' with file_list being "local_file@hostname:remote_file[,...]", but qsub from 2.5.10 won't allow such syntax. Since Torque is widely used by the European Grid middleware (http://repository.egi.eu/) and it used to rely on the old behaviour of -W, it will be good to restore that (though, just now it uses a workaround that sticks -W stagein=a@h:b,stagein=c@h:d' to the submission scripts). And the previous patch that allows for '-W stagein=a@h:b,stagein=c@h:d' is an absolute must: without it nothing will be submitted to the batch system by the EGI CREAM middleware.
Forgot to say: these patches are currently deployed at our production infrastructure and are OK already for some hours ;))
Found out that processing of double and single quotes and backslashes from smart_strtok was dropped in my patch. Will try to resurrect them and prepare the updated fix.
Fixed in 2.5.11 revision 5803
When submitting a job that contains: #PBS -W stagein=file1@host:/some/path/file1,stagein=file2@host:/some/path/file2 or: #PBS -W stagein=file1@host:/some/path/file1,file2@host:/some/path/file2 or: #PBS -W 'stagein=file1@host:/some/path/file1','stagein=file2@host:/some/path/file2' etc etc; We get the error: * qsub: illegal -W value However, submitting like this: #PBS -W stagein=file1@host:/some/path/file1 #PBS -W stagein=file2@host:/some/path/file2 It works fine. I believe 2.4 suffered from a similar bug, which was fixed in 2.4.9 (from changelog): b - restore ability to parse -W x=geometry:{...,...} Is it possible that this bug was not fixed in 2.5 ? Or maybe this is an completely different bug.